DeepSeek V2-0628Open Weight

Released

DeepSeek V2-0628 is an AI model released by DeepSeek on Friday, Jun 28 2024, 14 days after DeepSeek Coder V2. It is an open-weight model — the trained weights are available to download and run. It is a 236B parameterThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. model with a 128k token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover HumanEval, and MATH.

Benchmarks

Coding

HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better.
84.76%
#31 of 65

Reasoning & science

MATHReasoning — accuracy on the original competition mathematics test set
71.02%
#14 of 43

Compare DeepSeek V2-0628 with

DeepSeek V2-0628

About

DeepSeek V2-0628 updated the chat API on June 28, 2024. DeepSeek reported improvements in programming, mathematics and reasoning, alongside changes to role-playing behavior.

The checkpoint later became one of the inputs to V2.5, which combined its conversational abilities with Coder V2-0724. The scores here use the June announcement's HumanEval and MATH evaluations, preserving their original measurements.

Frequently asked questions

DeepSeek V2-0628 was released by DeepSeek on Friday, Jun 28 2024.

All DeepSeek releases

30 tracked